Add warning about the limitations of type inference when using Final

[sisp]

Unfortunately, I implemented type inference from a default value when using Final incorrectly in #150, i.e. it only works for simple cases, but it's actually not true type inference. I'm not sure whether it's currently possible to get the type hint inferred from the default value at runtime.

E.g.:

@dataclasses.dataclass
class A:
    data: Final = dataclasses.field(default_factory=lambda: [1, "a"])  # inferred as `list[int | str]`

1. The implementation in Add support for Final type #150 does not support dataclasses.field(...), but this could be fixed.
2. The implementation in Add support for Final type #150 would at best infer the default value to have the type list but cannot infer List[Union[int, str]]. In order to get the correct type, actual type inference is be needed.

This PR changes the behavior, so that a NotImplementedError is raised when Final has no argument and a default value exists. I don't think it's a big problem to not support this case because type inference is just a convenience. Instead, Final[...] could always be typed explicitly.

Sorry for the sloppiness in #150 ...

[lovasoa]

I think the behavior of #150 is ok... Of course you cannot get a complete type definition from just a single value, but using type can be useful in simple cases. Maybe just add a warning in this case ?

[sisp]

Good idea. I've pushed new commits. Something like this?

The commit message used for merging needs to be changed, perhaps you could use the commit message of 6393529.

[lovasoa]

Given this message, it will not be obvious where the issue comes from. I thing the warning should mention that it comes from marshmallow_dataclass, the name of the type, and the name of the field.

[sisp]

Agreed. I've pushed a new commit, but I don't know how to get the name of the field. Is there a way to get it from within field_for_schema(...)?

[sisp]

I've added a warning about the limitations of using a default factory according to our discussion in #152.

[lovasoa]

thanks and sorry for the delay !

[sisp]

No worries, thanks for merging the PR! 🙂